Healthcare (Commonwealth Union) – Scientists have created an AI-driven system that can identify, within just 10 seconds during surgery, whether any removable portions of a cancerous brain tumor remain, according to a study published in Nature.
The tool, named FastGlioma, significantly outperformed traditional techniques for detecting residual tumor tissue, as reported by researchers from the University of Michigan and the University of California, San Francisco.
Senior author Todd Hollon, M.D., a neurosurgeon at University of Michigan Health and an assistant professor at U-M Medical School pointed out that FastGlioma is an AI-powered diagnostic tool with the potential to revolutionize neurosurgery by enhancing the real-time management of patients with diffuse gliomas.
Hollon added that this technology is faster and more precise than current tumor detection methods and could be adapted for diagnosing other types of brain tumors in both adults and children. It lays the groundwork for advancing brain tumor surgical guidance.
When neurosurgeons take out a life-threatening tumor from a patient’s brain, they often cannot eliminate the entire tumor.
The portion left behind is referred to as the residual tumor.
This residual tissue frequently goes undetected during surgery because distinguishing between healthy brain tissue and tumor remnants in the surgical cavity is difficult.
Residual tumors can closely resemble normal brain tissue, posing a significant challenge in neurosurgery.
To identify residual tumor during operations, surgical teams use various techniques.
These include obtaining MRI scans, which require specialized intraoperative equipment that is not universally available, or using fluorescent imaging agents, which are effective only for certain tumor types.
These limitations restrict the broader application of these methods.
In a groundbreaking international study of AI-driven technology, neurosurgical teams examined fresh, unprocessed tissue samples from 220 patients who underwent surgery for low- or high-grade diffuse gliomas.
The AI tool, FastGlioma, detected and quantified residual tumor with an average accuracy of about 92%.
When comparing surgeries guided by FastGlioma to those using traditional imaging or fluorescent methods, the AI missed high-risk residual tumor in only 3.8 percent of cases, significantly outperforming the nearly 25 percent miss rate of conventional techniques.
“This model is an innovative departure from existing surgical techniques by rapidly identifying tumor infiltration at microscopic resolution using AI, greatly reducing the risk of missing residual tumor in the area where a glioma is resected,” explained co-senior author Shawn Hervey-Jumper, M.D., professor of neurosurgery at University of California San Francisco and a former neurosurgery resident at U-M Health.
“The development of FastGlioma can minimize the reliance on radiographic imaging, contrast enhancement or fluorescent labels to achieve maximal tumor removal.”
To evaluate the remnants of a brain tumor, FastGlioma integrates advanced optical imaging techniques with a category of artificial intelligence known as foundation models.
The models, consist of GPT-4 and DALL·E 3, are trained on vast and diverse datasets, making them highly adaptable for various applications.
Following extensive training, foundation models can perform tasks like image classification, functioning as chatbots, responding to emails, and creating images based on textual descriptions.
For FastGlioma, researchers worked on a visual foundation model using data from over 11,000 surgical specimens and 4 million distinct microscopic images.
The specimens were analyzed using stimulated Raman histology, a high-resolution optical imaging technique developed at the University of Michigan (U-M).
This same technology previously enabled the creation of DeepGlioma, an AI-powered diagnostic tool capable of identifying genetic mutations in brain tumors in less than 90 seconds.
